Development of a Machine Learning Algorithm for Drug Screening Analysis on High-Resolution UPLC-MSE/QTOF Mass...

Area of Science:

• Analytical Chemistry
• Forensic Science
• Biochemistry

Background:

• Ultra-performance liquid chromatography (UPLC)-MSE/quadrupole time-of-flight (QTOF) high-resolution mass spectrometry is used for untargeted drug screening.
• Systematic investigation of algorithmic analysis and positivity criteria for comprehensive drug screening using this method is lacking.
• The stability and utility of ion ratio (IR) in MSE/QTOF data analysis require further clarification.

Purpose of the Study:

• To systematically investigate the algorithmic analysis and positivity criteria for drug identification using UPLC-MSE/QTOF.
• To determine if ion ratio (IR) is a stable parameter for MSE/QTOF data analysis.
• To develop and validate a data-driven model for objective drug identification.

Main Methods:

• Experimentally determined IR for 91 drugs across varying concentrations and days.
• Employed a data-driven machine learning approach using multivariate linear regression (MLR).
• Incorporated parameters like mass error, retention time, fragment ion count, IR, isotope abundance accuracy, and peak response.

Main Results:

• Ion ratios were generally low and concentration-dependent for most compounds analyzed by MSE/QTOF.
• Developed an MLR model with composite scores using 7 parameters for positive drug identification.
• Achieved 89.38% mean accuracy in the validation set and 87.92% agreement in the test set.

Conclusions:

• The developed MLR model can serve as a decision-support tool for objective drug identification.
• Incorporating multiple parameters enhances the reliability of drug identification using UPLC-MSE/QTOF.
• This approach facilitates more objective and accurate drug screening.